Search

Childhood maltreatment may adversely affect brain development and consequently behavioral, emotional, and psychological patterns during adulthood. In this study, we propose an analytical pipeline for modeling the altered topological structure of brain white matter structure in maltreated and typically developing children. We perform topological data analysis (TDA) to assess the alteration in global topology of the brain white-matter structural covariance network for child participants. We use persistent homology, an algebraic technique in TDA, to analyze topological features in the brain covariance networks constructed from structural magnetic resonance imaging (MRI) and diffusion tensor imaging (DTI). We develop a novel framework for statistical inference based on the Wasserstein distance to assess the significance of the observed topological differences. Using these methods in comparing maltreated children to a typically developing sample, we find that maltreatment may increase homogeneity in white matter structures and thus induce higher correlations in the structural covariance; this is reflected in the topological profile. Our findings strongly demonstrates that TDA can be used as a baseline framework to model altered topological structures of the brain.

This thesis explores the use of computational chemistry and machine learning techniques to aid in the design of Metal-Organic Frameworks (MOFs) for use in postcombustion carbon capture and storage (PoC-CCS). PoC-CCS is an ongoing field of research which aims to selectively remove carbon dioxide, an important greenhouse gas, from the exhaust of fossil-fuel burning powerplants. By using a suite of advanced simulation techniques, high-throughput screenings were performed on thousands of MOFs to study their behaviour in a pressure swing adsorption (PSA) system. To develop a comprehensive picture of a material’s performance, the behaviour of individual gas molecules within the pores of the crystal structures to the material’s performance in industrial scale PSA columns was evaluated. To study the behaviour of individual gas molecules within the pores of a MOF, a new algorithm which can accurately determine the locations of gas binding sites was developed. This algorithm, which relies on probability distributions generated through grand canonical Monte Carlo simulations (GCMC), was optimized for CO2 with the goal of use in high-throughput screening. By tuning the user-controlled parameters for a desired gas, this algorithm, which was named the Guest Atom Localization Algorithm (GALA), was shown to accurately reproduce experimentally determined binding sites while being run in a high-throughput manner with no user intervention. Studying MOFs at the pore or crystal scale in this manner provides valuable insights into the behaviour of gases within the materials. A major shortcoming, however, is the lack of direct insight into the material’s behaviour in industrial systems. Materials scientists and MOF chemists have historically focused on a set of performance metrics measured at this scale; however, no clear connection can be made between such metrics and the performance of that sorbent material in a PSA column. To bridge this gap between MOF chemists and the process engineers studying the PSA systems, a large-scale screening of MOFs was performed using a sophisticated PSA simulator designed to reproduce the performance of an 80 kg PSA column. By supplying isotherms obtained using GCMC simulations to be used as inputs into the PSA simulator, a multi-scale high-throughput screening of MOFs for PoC-CCS was performed for the first time under coal-fired powerplant conditions. This multi-scale screening provided the ideal conditions to study the materials science performance metrics and their relationships to industrial PSA performance. To study this relationship, a series of machine learning and artificial intelligence techniques were employed. The primary goal was to extract important relationships between the materials science and industrial PSA performance metrics, with a secondary goal of developing a predictive model which could be used to accelerate the pace of materials discovery. Through the use of machine learning, several metrics were identified which could be used to predict whether a material could meet the minimum target of 95 % purity of captured CO2, and 90 % removal (or recovery) of CO2 from the flue gas stream. Among them was the isotherm parameters for N2, the most abundant species in the flue gas. This finding was significant as to date the focus among MOF chemists studying the PoC-CCS system was placed primarily on the CO2 metrics, with N2 only implicitly considered when calculating the CO2/N2 selectivity. Although several metrics were identified which could predict the purity and recovery targets, none of the conventional metrics tested could be used to estimate the energetic cost of capture or the size of the capture plant, both important considerations in evaluating the cost of capture. The relationship between N2 binding within the pores of the MOF and its ability to meet the purity-recovery targets was explored using GALA. Using a Tanimoto similarity metric and the ratio of single component and competitive loadings, the CO2 and N2 binding environments were studied. It was determined that when the N2 binding environment was significantly altered by the presence of CO2, the material was more likely to meet the purity-recovery targets. Further analysis found that this change in binding environments was correlated to a reduced N2 uptake in the presence of CO2, implying that the competition for binding sites within the pores of the MOF is an important indicator for the material’s ability to meet the purity-recovery target. For the first time, a direct relationship between the behaviour of individual gas molecules to industrial PSA performance can be reported. Although the PSA simulator used throughout this work has proven to be a powerful tool for materials discovery, several shortcomings still exist. The first is the method used by the simulator to predict the loadings at various points within the column. This method relies on single component isotherm data despite the ability of GCMC to simulate multi-component isotherms. An alternative method to using single component isotherms was proposed which relies on multi-component isotherm data and a linear interpolation model. The existing method was compared to the new proposed interpolation method, and it was found that the loadings predicted using the interpolation method were more accurate. The second shortcoming of the PSA simulator is the computational expense associated with the optimizations. Using the PSA simulator, a single material may take up to a week to be fully optimized on a high-performance computing cluster. To increase the pace of materials discovery, a surrogate model was developed using the data accumulated over the course of the work presented in this thesis. Using artificial neural networks, a suite of models was developed which reproduces the outputs of the PSA simulator and is able to optimize a single MOF in a matter of minutes. This suite of models, known as the Fossil Fuel Combustion for Carbon Capture and Storage (FoCAS) was used to perform a screening of over 4,000 materials.